Speech Recognition Channel: Faster-Whisper-XXL

What is this?

Faster-Whisper-XXL is a standalone executable (faster-whisper-xxl.exe) provided by the Purfview/whisper-standalone-win project. It packages the faster-whisper speech recognition engine into a single exe file that can be used without installing a Python environment.

pyVideoTrans has built-in integration for this tool. After selecting this channel, you can directly invoke faster-whisper-xxl.exe for speech recognition.

Use Cases

• Users who don't want to install Python and prefer a plug-and-play experience
• Users who have already downloaded faster-whisper-xxl.exe and want to use it directly in pyVideoTrans
• Users who need offline recognition with high accuracy
• Users with an NVIDIA GPU for the best performance

Prerequisites

Requirement	Details
faster-whisper-xxl.exe	Must be downloaded from GitHub
NVIDIA GPU (recommended)	Supports CUDA acceleration with significant speed improvement; CPU mode available without GPU
FFmpeg (optional)	May be needed for some audio formats

Download

Download faster-whisper-xxl.exe from:

https://github.com/Purfview/whisper-standalone-win/releases

Extract to any folder (path must not contain Chinese characters or spaces).

Using in pyVideoTrans

Configuration Steps

1. Download and extract faster-whisper-xxl.exe
2. Open pyVideoTrans
3. In the speech recognition channel, select faster-whisper-xxl.exe
4. In the file picker, locate faster-whisper-xxl.exe
5. Select it and you're ready to use

Parameter Reference

When pyVideoTrans calls faster-whisper-xxl.exe, it automatically passes the following parameters:

Parameter	Description
Audio file path	The audio file to recognize
-pp	Pretty Print — formatted output
-f srt	Output format as SRT subtitles
-ct	Compute type (default int8, configurable in settings)
-l	Recognition language code
--model	Model name to use
--output_dir	Output directory (cache folder)
--initial_prompt	Initial prompt (optional, improves domain-specific recognition)

Custom Additional Parameters

Tip: faster-whisper-xxl.exe supports many parameters. For flexibility, the software only includes the essential model, language, and output format parameters. You can add additional parameters by creating a file called pyvideotrans.txt in the same directory as faster-whisper-xxl.exe.

Note: Do not duplicate the model size, language, or output format parameters, as this will cause errors.

For example, create pyvideotrans.txt in the same directory as faster-whisper-xxl.exe:

The file contents should be the additional parameters you want to add, separated by spaces.

Optimal Configuration

Setting	Recommended Value	Notes
Model	large-v3 or large-v3-turbo	Highest accuracy, suitable for most scenarios
Compute type (ct)	int8	Balanced speed and accuracy; use float16 with a GPU
Language	Specify the language	More accurate than auto-detection
Initial prompt	Domain-related terminology	Improves recognition accuracy for specialized vocabulary
Compute device	GPU (CUDA)	Strongly recommended with NVIDIA GPU

Common Issues

Q: "File not found" after selecting the exe?

A: Ensure the path is correct and does not contain Chinese characters or spaces. Place the exe in a pure-English path.

Q: Recognition is very slow?

A:

• Check if GPU mode is enabled (requires NVIDIA GPU and CUDA support)
• Try switching compute type to int8 (faster)
• Use a smaller model (e.g., base or small) for faster speed

Q: Recognition results are inaccurate?

A:

• Try a larger model (e.g., large-v3)
• Set the correct language parameter
• Add an initial prompt containing domain terms or names from the audio

Q: Parameters in pyvideotrans.txt don't take effect?

A: Ensure the file is in the same directory as faster-whisper-xxl.exe, not in the pyVideoTrans installation directory. Parameters should be separated by spaces.

Q: How to check which models are supported?

A: Visit https://github.com/Purfview/whisper-standalone-win/releases for project documentation. Supported models are the same as faster-whisper.